Renewable Energy 37 (2012) 285e298

Contents lists available at ScienceDirect

Renewable Energy
journal homepage: www.elsevier.com/locate/renene

Dynamic overset CFD simulations of wind turbine aerodynamics

Yuwei Li a,1, Kwang-Jun Paik a, 2, Tao Xing b, 3, Pablo M. Carrica c, *
a
IIHR-Hydroscience and Engineering, C. Maxwell Stanley Hydraulics Laboratory, The University of Iowa, Iowa City, IA 52242, USA
b
Department of Mechanical Engineering, College of Engineering, University of Idaho, Moscow, ID 83844, USA
c
Department of Mechanical and Industrial Engineering and IIHR-Hydroscience and Engineering, C. Maxwell Stanley Hydraulics Laboratory, The University of Iowa, Iowa City,
IA 52242, USA

a r t i c l e i n f o a b s t r a c t

Article history: Simulations of the National Renewable Energy Laboratory (NREL) phase VI wind turbine using dynamic
Received 21 November 2010 overset grid technology are presented. The simulations are performed in an inertial frame of reference with
Accepted 18 June 2011 the rotor consisting of the blades and hub. The geometries of the tower and nacelle are approximate but
Available online 18 July 2011
included in the computation. Computations of the effect of wind speed (5, 10, 15 and 25 m/s) at a xed blade
pitch angle of 3
with constant rotational speed using unsteady Reynolds-Averaged NaviereStokes (RANS)
Keywords:
and Detached Eddy Simulation (DES) turbulence models, both showing little difference in the averaged
Wind turbine aerodynamics
forces and moments. However, signicant improvements in the transient response are seen when using
Overset grids
Rotor ow
DES. The effect of angle of attack is evaluated by dynamically changing the pitch from 15
to 40
at
Computational uid dynamics constant wind speed of 15 m/s. Extensive comparison against experimental results, including total power
and thrust, sectional performance of normal force coefcient and local pressure coefcient, shows
consistently good predictions. The methodology shows a promise for more complex computations
including active turbine control by varying the pitch angle and uid-structure interaction.

2011 Elsevier Ltd. All rights reserved.

1. Introduction
Wind energy available around the world is much greater than
the current world energy consumption. The generation potential of
wind power on land and near off-shore is estimated at 72 TW, over
ve times the worlds current energy use in all forms [1]. Predic-
tions show that the global electric wind capacity will stand at
409 GW in 2014, up from 158 GW at the end of 2008, with an
average annual growth rate of 20.9% [2]. Horizontal wind turbines
are the least expensive and clean way to harness this important
energy source. However, for the design and development of more
efcient and reliable wind turbines, accurate prediction of aero-
dynamic behavior is of critical signicance, since the interaction of
the wind with the blades inuences the efciency. It also has
a signicant effect on the loads on bearings and gearbox, ultimately
affecting the lifespan and reliability of the machine.
The ow in wind turbines, even in very large ones, is still
essentially incompressible, with Mach numbers based on blade tip
* Corresponding author. Tel.: 1 319 335 6381; fax: 1 319 335 5238.
E-mail addresses: yuwei-li@uiowa.edu (Y. Li), kwangpaik.paik@samsung.com
(K.-J. Paik), xingtao@gmail.com (T. Xing), pablo-carrica@uiowa.edu (P.M. Carrica).
1 the more advanced and costly Detached Eddy Simulation (DES)
Tel.: 1 319 855 3279; fax: 1 319 335 5238.
2
Present address: Samsung Heavy Industries, Daejeon, South Korea.
3
Tel.: 1 208 885 6579 (ofce); fax: 1 208 885 9031.
speed rarely exceeding 0.25. This fact justies the use of incom-
pressible uid solvers for most wind turbines. Methods of various
levels of complexity to predict the aerodynamic behavior of a wind
turbine rotor have been developed. Being computationally cheap
and highly efcient, blade element momentum methods (BEM)
have been very popular for engineering design, provided that good
airfoil data are available for lift and drag coefcients as a function of
angle of attack [3]. Several codes and models were developed based
on BEM and their performance was improved by introducing new
correction models such as tip loss and dynamic stall corrections
[4,5]. However, BEM models are greatly inuenced by the choice of
airfoil data and dependent on empirical corrections to two-
dimensional (2D) airfoil results to account for three-dimensional
(3D) effects, such as tip loss, rotational ow, and dynamic stall
[6]. To obtain more physics of wind turbine aerodynamics and
retain high computational efciency, 3D inviscid aerodynamic
models, in which viscous effects are neglected, were introduced,
including lifting line [7], panel [8], vortex [9], and Boundary Integral
Equation (BIEM) [10] methods. Nevertheless, issues arise because
potential ow methods cannot handle viscous effects and separa-
tion. The next level of complexity is to solve the Reynolds-Averaged
NaviereStokes (RANS) equations with some turbulence models, or
approaches that combine the accuracy of large eddy simulation
(LES) inside the separation region for vortical structures and

0960-1481/$ e see front matter
2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.renene.2011.06.029
286 Y. Li et al. / Renewable Energy 37 (2012) 285e298
efciency of RANS inside a boundary layer. This gives DES the ability
to better resolve ow separation and the stall of the airfoil [11].
Though the cost of these approaches is signicantly higher than any
of the previously mentioned simpler methods, advances in
computer technology make it possible to handle large, dynamic
problems with parallel platforms.
Most computations to date testing numerical methods are
compared to the National Renewable Energy Laboratory (NREL)
Unsteady Aerodynamics Experiment (UAE) [12,13], which provides
comprehensive high-quality data for a modied Grumman 20 kW
twin-bladed turbine, tested in the wind tunnel at NASA Ames. The
most important results are in the Phase VI of the experiments. A
blind numerical study involving 20 different participants using
various CFD codes was conducted subsequently [14].
Several authors have performed CFD computations of wind
turbines with a variety of methods. Sezer-Uzol and Long [15]
computed the NREL Phase VI turbine at different wind speeds and
yaw angles using the nite volume ow solver PUMA2 with rotating
unstructured tetrahedral grids, showing good agreement with
experiment, but the inviscid nature of the code resulted in limited
ability to predict situations when massive ow separation occurs.
Srensen et al. [16] studied 3-D aerodynamic effects as a function of
wind speeds by using the multiblock nite volume, incompressible
RANS ow solver EllipSys3D with a rotor-only conguration. Good
qualitative and quantitative agreement with experimental
measurements evidenced the advantages of CFD approaches for
wind turbine simulation. Perhaps the most comprehensive aero-
dynamic study to date has been performed by Duque et al. [17], who
performed computations of the NREL Phase VI turbine with the
NASA compressible RANS ow solver Overow-D, based on a nite
differences approach and overset grid [18]. The authors compare the
results of Overow-D and the lifting line code CAMRAD II with the
experiments, and extensively discuss the aerodynamic performance
of the wind turbine, including shaft power, normal force and pres-
sure coefcient. Potsdam and Mavriplis [19] used the unstructured
multigrid RANS code NSU3D to predict the aerodynamics of an
isolated wind turbine rotor, and the results were compared with
both the experiments and the predictions with the code Overow.
Bazilevs et al. [20] studied the rotor of the NREL 5 MW baseline wind
turbine [21] using both a nite element approach and a NURB-based
(Non-Uniform Rational B-splines) approach for the geometry, which
has the potential for coupled aerodynamic/structural analysis.
Notice that most studies consider the rotor-only geometry,
excluding the tower and nacelle; in most cases only one blade was
included in the simulation. These simplications are understand-
able because the resulting grid is static throughout the calculation
and thus greatly alleviate the computational complexity and cost,
but they can leave out some important effects. Dynamic moving
overset grids is one of the methods that allow for computation of
bodies with relative motions, including elastic deformation.
The objective of this paper is to present transient computations
of the full-scale NREL Phase VI turbine. The study is performed
using the incompressible, dynamic overset code CFDShip-Iowa v4.5
with a hierarchy of objects that include the blades, rotor, nacelle,
and tower/ground, allowing variations of blade pitch angle and yaw
during the computations. The tests include cases with a xed blade
pitch angle (3
) and variable wind speeds, and variable pitch angles
at xed wind speed (15 m/s). All tests are performed at a constant
rotational speed of 72 RPM. For these cases extensive comparison
with experimental data is performed and the results are analyzed.
2. Mathematical and numerical methods
The general purpose code CFDShip-Iowa v4.5 is used to perform
the wind turbine computations. CFDShip-Iowa v4.5 is a nite
difference, general-purpose unsteady Reynolds-Averaged Navier-
Stokes (URANS) or Detached Eddy Simulation (DES) overset solver.
The air side is treated with a semi-coupled approach, in which the
water ignores the presence of the air but the air is computed using
the water free surface as an immerse boundary, thus providing an
excellent approximation to the air ow and forces on large-scale
objects like ships or oating structures [22]. The free surface is
modeled with a level set approach, enforcing kinematic and
dynamic free surface boundary conditions on the interface [23].
Dynamic overset grids are used to resolve grid deformation and
relative motions [24], where the overset connectivity is provided at
run time by the code Suggar [25], which allows dynamic interpo-
lations. The code and the overset strategy have recently been
optimized for large-scale computations [26]. The code has capa-
bilities for full six degree of freedom (6DOF) and a parent/child
hierarchy of objects that allows motion of control surfaces and
other appendages. Autopilots based on proportional-integral-
derivative (PID) controllers allow control of heading, speed or
attitude. Fluid-structure interaction (FSI) can also be modeled by
using either modal superposition for linear problems or a structure
solver based on nite element method for non-linear problems
[27]. Convection terms are discretized with nite differences
second-order upwind (for RANS) or fourth-order upwind biased
(for DES or delayed DES), and with a second-order centered scheme
for the viscous terms. The temporal terms are discretized using
a second-order backwards Euler scheme. Incompressibility is
enforced by a strong pressure/velocity coupling, achieved using
either the pressure implicit with splitting of operators (PISO) or
projection algorithms.
Specic discussion follows on terms related to the computation
of the wind turbine. For complete details on the mathematical and
numerical methods the reader is referred to the cited references
and the literature therein.
2.1. Governing equations
Mass and momentum conservation equations are written in
dimensionless form as follows:
V,u 0 (1)
" #
vu 1  
u,Vu Vp V, Vu VuT S (2)
vt Reeff
where u is the uid velocity and S is a source term, zero in this
paper. p is the non-dimensional pressure, Reeff is the effective
Reynolds number, dened as:
pabs 2
p k (3)
rU02 3
U0 L
Reeff (4)
v vt
where pabs is the absolute pressure, U0 and L are the free-stream
velocity and characteristic length (in this case the radius of the
blade) respectively, nt is the turbulent eddy viscosity, and k is the
turbulent kinetic energy.
2.2. Turbulence modeling
The turbulence is modeled using a blended ku/k3 shear stress
transport (SST) model [28], in which the turbulent kinetic energy k
and specic dissipation rate u are
 Y. Li et al. / Renewable Energy 37 (2012) 285e298 287

vk 1 and 25 m/s) at a xed 3
blade tip pitch angle. Simulations for

u  sk Vyt ,Vk  V2 k sk 0 (5) sequence K are designed to evaluate the performance of the 3D
vt Pk
blade in the presence of rotation to different static angles of attack
by varying the blade tip pitch angles from 40
to 15
in steps of 5
.
vu 1 The wind velocity for sequence K is xed at 15 m/s (the experi-
u  su Vyt ,Vu  V2 u su 0 (6)
vt Pu mental data also evaluates 6, 10 and 20 m/s). For both cases the
where the turbulent viscosity and effective Peclet numbers are rotational speed is 72 RPM. Several yaw angles were tested exper-
dened as imentally, but only those with the yaw angle xed at 0
were
simulated. Table 1 summarizes all CFD simulation cases.
k 1 1 RANS and DES computations were conducted for sequence S to
yt ; Pk ; Pu (7)
u 1 1 investigate the response of the two methodologies for wind turbine
sk yt su yt modeling, while only DES was applied to sequence K. In both cases
Re Re
hybrid 2nde4th order schemes were used for convection.
and the source terms for k and u are
The grid design for sequences S and K is shown in Fig. 1. The
geometries of the nacelle and hub have been approximated and do
k3=2
sk G (8) not correspond to the true shape of the NREL phase VI turbine. The
lku
grid system consists of 13 overset blocks to discretize the blades,
hub, nacelle, tower and oor. The blades themselves are each
u * 1
su  g G b u2  21  F1 su2 Vk$Vu (9) gridded with an O-type block to cover the span, one block for the
k u tip and one for the root. O blocks are used for the hub and nacelle as
p *
where the length scale is lku k=b u and the kinetic energy well as for the tower. A renement block is used to match the
production is G yt s : Vu. The blending function that switches nacelle grid with those from the hub/blade roots/tower systems.
between the ku model near the wall to the k3 model on the free Finer renements are used to capture the ow around the rotor
stream region is using a Cartesian block and an O-type block. Overall 52.3 million
" p ! !!4 # grid points are used, distributed in 2048 domains each sent to
k 1 500 4su2 k a processor, with an average of 28 046 grid points per processor and
F1 tanh min max ; ; (10) a maximum deviation around the average of 2.2% maximum. After
0:09ud Re d2 u CD u d2
k
splitting, the 52.3 million grid points increase to 57.4 million due to
where d is the distance to the nearest no slip surface and duplication on the block to block interfaces, see Table 2.
1 Since surface overset grids are used, the computation of forces
CDku max2su2 Vk,Vu; 1020 . This model takes advantage of
u and areas requires evaluation of the portions that are overset to
the benets of the ku model, e.g. does not require near-wall avoid double-counting in overlaid areas. This is done as a pre-
damping function and uses simple Dirichlet boundary conditions processing step with the code Usurp [31], which generates
at solid walls, and of the k3, e.g. less sensitive to the level of free- weights that alter the area of each cell on the solid body to provide
stream turbulence. the appropriate forces and areas at each cell.
DES modeling follows Travin et al [29], where the length scale is The grids are organized in a parent/child hierarchy, as shown in
computed as Table 2. The grids comprising the blades can pitch about the
pitching axis, while these and the hub form the rotor, which rotates
~l minl
ku ; CDES D (11) about the shaft axis. The nacelle and the rotor form the body
where D is the grid length scale, taken as the maximum cell length Nacelle which can rotate about the tower axis with the yaw angle.
side, and In this way a fully controlled turbine can be simulated, with
a rotational speed controller acting on the blade pitch and an
k
CDES 1  F1 CDES ku
1  F1 CDES (12) attitude controller acting on the yaw. In this work the yaw is kept at
0
, but the pitch is modied dynamically to perform the simula-
k 0:61 and C ku 0:78. Delayed-DES (DDES) modeling
with CDES DES tions of sequence K. The time step was chosen such that the blades
in CFDShip-Iowa follows the approach of Sainte-Rose et al. [30]. rotate 1
per time step for cases with 5e15 m/s wind speed, and
For the simulations presented in this paper, integration all the 0.5
for cases with 25 m/s wind speed.
way to the wall is used (no wall functions). The SST model then
requires wall renement to satisfy y  1. 4. Results and discussion

3. NREL phase VI turbine and grid strategy 4.1. Variable wind speed at constant pitch angle

The testing wind turbine is the NREL phase VI, a modied
Grumman Windstream 33 stall-regulated turbine with full-span
pitch control and a power rating of 20 kW. It has 2 blades, with
NREL s809 tapered and twisted blade prole. The rotor diameter is
10.058 m while hub height is 12.192 m. The experiments were
performed in the NASA Ames wind tunnel in 1999 and are
considered a benchmark for evaluation of wind turbine aero-
dynamics computer codes. Detailed geometry, machine parameters
and experimental procedures can be found in the NREL report [12].
Two cases were selected from the test matrix of the NREL
experiments, belonging to sequences S and K. In particular, simu-
lations for sequence S are used to evaluate the ability of the code to
predict the aerodynamics under different wind velocities (5, 10, 15
The experimental sequence S comprises data at 3
of blade pitch
for wind speeds from 5 to 25 m/s at intervals of 1 m/s. The CFD
study is focused on 5, 10, 15 and 25 m/s. The highest two wind
speeds correspond to stall conditions in most of the blade and
simulations are thus challenging. This is shown in Fig. 2, which
illustrates the vortical structures using iso-surfaces of the second
invariant of the rate of strain tensor [32] at Q 5. It is clear that
these DES computations predict fully attached ow for 5 and 10 m/
s, with development of unsteady trailing vortices at 10 m/s. At these
two velocities the blade tip vortices are strong and stable. Strong
vortices detach also from the tower and the roots of the blades,
where the geometry changes quickly from the s809 prole to
cylindrical posts attached to the hub. Notice that the vortical
288 Y. Li et al. / Renewable Energy 37 (2012) 285e298

Table 1 structures as the wind speed increases. The formation of these

Simulation cases. hairpin vortices is interesting, since it requires the presence of
Case Nominal Re Pitch Run RANS/DES gradients of streamwise velocities [33], which in this case occur
no. wind speed (based on [degree] along the rotor, and especially by the rotor tip vortex.
[m/s] blade length) Supplementary video related to this article can be found at doi:
1 5 1.766  106 3 S0500000 RANS 10.1016/j.renene.2011.06.029
2 15 5.205  106 3 S1500000 RANS
Fig. 3 shows comparisons of predicted vortical structures at two
3 25 8.645  106 3 S2500001 RANS
4 5 1.766  106 3 S0500000 DES wind speeds for RANS and DES computations. At 5 m/s both
5 10 3.535  106 3 S1000000 DES methods predict essentially the same ow around the blades. More
6 15 5.205  106 3 S1500000 DES vortices can be observed for DES in the separated ow regions close
7 25 8.645  106 3 S2500001 DES to the hub, tower and nacelle, but these have lesser effect on the
8 15 5.241  106 15 to 40 K1500ST1 DES
in 5 degree
performance of the turbine. At wind speed of 25 m/s the differences
steps between RANS and DES are more dramatic. In particular the
separation bubbles on the suction side of the blade are highly
unsteady and are shed periodically for the DES computation, while
structures dissipate quickly away from the regions covered by grid they are mostly steady for RANS. This causes a stable tip vortex for
renements, downstream of the rotor plane and at the tower below RANS, while the tip vortex breaks down for DES, as previously
1 blade length. At 15 m/s about the outer half of the blade experi- discussed. Animations of RANS simulations in the earth reference
ences massive unsteady separation, while at 25 m/s most of the frame are shown in Electronic Annexes VII and VIII for 5 and 25 m/s,
blade suffers massive unsteady ow separation. For these two respectively. Notice in Electronic Annexes IV and VIII that the
higher velocities the interaction of the tip vortices with the separation bubble is essentially steady in RANS but highly unsteady
unsteady separation from the suction side of the blades causes with DES.
breakdown of the tip vortices, very dramatically at 25 m/s. Elec- The experimental thrust and torque (and thus power) are
tronic Annexes I, II, III and IV show animations of the vortical obtained integrating the pressure measurements along the blade,
structures colored with axial speed obtained with DES for wind and consequently the friction effects are neglected. CFD computa-
speeds of 5, 10, 15 and 25 m/s, respectively, in a blade-xed refer- tions account for both pressure and friction forces. Figs. 4 and 5
ence frame. The unsteadiness and regions of attached or separated show comparisons of thrust and power between DES-based CFD
ow on the blades under different conditions are clear in these and NREL experiments for Sequence S. In all gures the vertical bars
animations. Animations of DES simulations at 5 and 25 m/s on the represent the experimental standard deviation, not the error or
earth reference frame are shown in Electronic Annexes V and VI, experimental uncertainty, which is not reported in [13]. Overall CFD
respectively. The animations show the separation on the tower and simulations predict very well the general performance of the
nacelle in a more transparent way. Note how the tower vertical turbine, even at the two highest wind speeds where stalled ows
vortices tend to evolve into hairpin vortices and into smaller occur. CFD predictions of the thrust are all well within the standard

Fig. 1. Grid design. Grid points are skipped in all directions for clarity.
 Y. Li et al. / Renewable Energy 37 (2012) 285e298 289

Table 2
Grid details.

Name imax jmax kmax Procs isplit jsplit ksplit Processor Total Hierarchy
points
Hub 121 101 51 24 41 26 26 27716 665K Rotor-Nacelle
Blade 1 201 101 241 192 35 26 31 28210 5.41M Blade-Rotor-Nacelle
Tip 1 121 101 101 48 41 26 26 27716 1.33M Blade-Rotor-Nacelle
Root 1 201 101 31 24 35 26 31 28210 677K Blade-Rotor-Nacelle
Blade 2 201 101 241 192 35 26 31 28210 5.41M Blade-Rotor-Nacelle
Tip 2 121 101 101 48 41 26 26 27716 1.33M Blade-Rotor-Nacelle
Root 2 201 101 31 24 35 26 31 28210 677K Blade-Rotor-Nacelle
Renement 241 262 262 648 31 30 30 27900 18.08M Earth
Ref. Tip 61 1081 201 512 31 35 26 28210 14.44M Nacelle
Ref. Nacelle 221 69 69 40 23 35 35 28175 1.13M Nacelle
Tower 214 61 71 36 37 31 25 28675 1.03M Earth
Nacelle 151 61 101 36 26 31 35 28210 1.02M Nacelle
Background 241 151 151 224 31 39 23 27807 6.23M Earth
Total 2048 Ave: 28046 57.43M

Fig. 2. Vortical structures represented by iso-surfaces of Q 5 for different velocities (pitch angle 3
).
290 Y. Li et al. / Renewable Energy 37 (2012) 285e298

Fig. 3. Comparison of vortical structures predicted by RANS and DES for low (5 m/s) and high (25 m/s) wind speeds. Vortical structures are represented by iso-surfaces of Q 5
(pitch angle 3
).

Fig. 4. Thrust for different velocities (pitch angle 3
). Fig. 5. Power to the shaft for different velocities (pitch angle 3
).
 Y. Li et al. / Renewable Energy 37 (2012) 285e298
Fig. 6. Normal force coefcients for different velocities at 5 radial sections (pitch angle
3
).
deviation of the experimental measurements, except for a slight
over prediction at 25 m/s where as previously shown stalled ow
and separation are pronounced. The friction component on the
total thrust is negligible. Power is a bit underpredicted, except for
the highest speed. In addition, results neglecting friction in the
computation of the forces are closer to the experimental data as
expected, but the friction force contribution is still small except at
Fig. 7. Pressure coefcient for different velocities on 5 blade sections (pitch angle 3
).
291
slower wind speeds. Notice that the code predicts properly the
attening of the power as a function of the wind speed as the
turbine becomes stall-controlled at higher wind speeds.
As thrust and power are parameters integrated over the area of
the blades, evaluation of sectional force coefcients such as radial
normal force coefcient Cn allows a better check on the ability of
the code to properly capture the aerodynamic behavior of the wind
turbine. The normal force coefcient is still an integration of
pressure but limited to a section of the blade. As pointed out by
Simms et al. [34], who summarized blind predictions of several
codes for the Phase VI turbine, good prediction of integrated
parameters can be obtained with models that over predict aero-
dynamic forces on the inboard part of the blade, while under pre-
dicting the forces outboard. Fig. 6 compares RANS and DES CFD and
experimental results of Cn at 5 different sections of the blade. Both
RANS and DES match very well the experimental measurements for
all wind velocities simulated. In particular, at lower wind velocities
(5 m/s) where no ow separation occurs except at the transition
section near the root, RANS and DES simulations show accurate
predictions with little difference between each other, as expected
from the similarity shown in the ow regimes from Fig. 3. At higher
wind velocities, where ow separation becomes more important
and vortex shedding occurs, discrepancies appear gradually. At
15 m/s RANS and DES results are very similar for r/R < 0.5 where
the ow around the blade separates weakly (see Fig. 2 and
292 Y. Li et al. / Renewable Energy 37 (2012) 285e298
Fig. 8. Limiting streamlines on the suction side of the blade and selected sections (colored with pressure) for 3
pitch angle.
Electronic Annex III for the DES results), but differ more for r/R > 0.5
where separation and vortex shedding are important and differ-
ences in ow pattern between RANS and DES are remarkable. At
25 m/s the differences between RANS and DES are signicant for all
sections, as expected from the dramatic difference in ow patterns
shown in Fig. 3. Observe that RANS computations tend to predict
more lift as the ow remains attached at larger angles of attack. All
normal force coefcients are well predicted with DES with the
exception of section r/R 0.63 at V 15 m/s and section r/R 0.47
at V 25 m/s. Large differences in normal force coefcients at these
sections and wind speeds are also shown in the computations of
Duque et al. [17].
Direct comparisons of the pressure coefcient CP between CFD
and experimental data at 5 different sections as a function of wind
speed are shown in Fig. 7. The pressure coefcient is a harder test
for CFD since it is a local quantity, and comparison is made against
pressure taps installed in the blade. See that the incompressible
formulation of CFDShip-Iowa, appropriate for wind turbines, does
not produce the spurious pressure peaks observed on compressible
codes [17]. At low wind speeds the CFD predictions match the
experimental data remarkably well. For 10 m/s at r/R 0.47 the
experiments show a at pressure on the suction side, indicating
separation, while CFD results predict a leading edge peak. The same
behavior occurs for 15 m/s at r/R 0.30, where CFD predicts a peak
at 0.2 chord lengths downstream of the leading edge while
experimental results show a more at pressure distribution. Notice
that the integral of the pressure for these two anomalous condi-
tions will be about the same for CFD and experiments, resulting in
good prediction of the normal force coefcient. Duque et al. [17]
argued that possibly unsteadiness is the reason for these discrep-
ancies, since they used the code Overow-D in steady-state mode.
This is not supported by the current simulations as similar differ-
ences are observed for DES that captures signicant unsteadiness in
the trailing edge on the suction side both for 10 m/s at r/R 0.47
and for 15 m/s at r/R 0.30 (see Electronic Annexes II and III). It is
possible that CFD grossly under-predicts the separation, though it is
hard to think of a mechanism that would separate locally at r/
R 0.47 for V 10 m/s while all other sections inboard and
outboard are attached. The two points in Fig. 6 that show the largest
errors, section r/R 0.63 at V 15 m/s and section r/R 0.47 at
 Y. Li et al. / Renewable Energy 37 (2012) 285e298 293

with the blade down at zero rotations. The measurements were

taken at ve different points on sections r/R 0.30, 0.47, 0.63, 0.80
and 0.95 at 44% of the chord from the leading edge on the pressure
side. The CFD values correspond to one rotation saving every 4
(5e15 m/s) or 8 (25 m/s) time steps (90 points per rotation) and are
repeated three times, due to the cost of saving large number of
volume solutions and the corresponding processing. Because the
absolute pressures are small, the differences between the mean
values for experiments and CFD appear large for all sections except
r/R 0.47, but the largest error occurs at r/R 0.30 and is less than
2.5% of the pressure dynamic range at that section, see Fig. 7.
Notice in Fig. 9b that the amplitude of the uctuations is in
general larger in the experiments than in CFD. At r/R 0.30 the
presence of the tower is clear, shown in Fig. 9a as an increase in
pressure at integer rotations (0, 1, 2, etc.), and in Fig. 9b as a peak at
1.2 Hz, the rotational frequency. Large amplitudes at 1.2 Hz and
2.4 Hz are present at all sections, but the amplitude decreases for
larger radii. Most of the frequency content is limited to frequencies
below 10 Hz, as can be expected for a machine of this size. In CFD
these high frequencies can be associated with vortex shedding (see
Electronic Annex III to observe the phenomena). The magnitude of
the uctuations is larger in the experiments than in CFD at the
other sections, but the difference is more evident at r/R 0.63.

Fig. 9. Cp on ve sections for 15 m/s, 3
pitch: (a) time histories, (b) FFT.

V 25 m/s, show signicant under prediction of the suction

pressure and thus result in lower integral inside the curves.
Fig. 8 shows instantaneous limiting streamlines on the blade,
along with streamlines on the projected velocities at the selected
sections. At low wind speed (5 m/s) the CFD computations predict
fully attached ow everywhere in the active blade, and some
separation where the s809 blade prole merges to a cylindrical
section. At this speed the pressure stays low on the suction side and
the blade provides consistent lift. At 15 m/s the ow exhibits
signicant separation at all sections, with smaller scale, unsteady
vortices shed in the second half of the blade (see also Electronic
Annex III). For section r/R 0.47 at this speed the ow separates
from the leading edge and reattaches to the blade surface to form
a closed separation bubble whereas all other sections show open
separation where vortices are shedding away from the blade
surface to the wake. Signicant pressure recovery is observed,
mostly for the outer sections. Similar trends are observed for all
sections at 25 m/s, but the separation is stronger with violent
vortex shedding, with very little pressure recovery on the suction
side indicating massive stall (see also Electronic Annex IV).
Time histories of Cp for 15 m/s at 3
blade pitch angle are shown
in Fig. 9a, and the corresponding frequency spectra are shown in
Fig. 9b. The time evolution is expressed in terms of blade rotations Fig. 10. Thrust for different velocities, 3
pitch: (a) time history, (b) FFT.
294 Y. Li et al. / Renewable Energy 37 (2012) 285e298

Fig. 11. Vortical structures represented by iso-surfaces of Q 5 for different pitch angles (wind velocity 15 m/s).

Notice that the frequency content in CFD is comparable with the
experiments, indicating that the turbulence model is able to
capture most of the oscillations caused by organized vortical
structures and uctuations caused by turbulent structures. On the
same line of analysis, the turbulence model seems to fail to capture
the amplitude of the pressure uctuations, likely due to insufcient
grid discretization to maintain the low pressure on the vortex cores,
though other causes may be speculated that would cause differ-
ences in pressure uctuation amplitudes, like blade vibrations
caused by gears and bearings and by the elasticity of the blade. All
these effects are neglected in CFD.
Time histories of thrust for 5, 10, 15 and 25 m/s at 3
blade pitch
angle are shown in Fig. 10a, and the frequency spectra in Fig. 10b. In
this case the forces are saved every time step, so there are 1080
points in CFD for 5e15 m/s and 2160 for 25 m/s. To compare with
the experiments fairly, the thrust in one blade is multiplied by two
instead of using the thrust in both blades, which would smooth the
transient behavior of CFD by adding two blades that have different
instantaneous forces. The experimental thrust was obtained by
integrating the pressures measured on the ve sections at
r/R 0.30, 0.47, 0.63, 0.80 and 0.95, while CFD forces are integrated
on every grid cell on the blade, smoothing transients more than in
 Y. Li et al. / Renewable Energy 37 (2012) 285e298
the experimental evaluation of the force by averaging local pressure
highs with local pressure lows. It is not surprising then that the
experiments show larger uctuation amplitudes than CFD, though
a signicant portion of the larger uctuations is likely due to
inadequacy of the turbulence model to fully capture the transient
instabilities observed in the ow, as discussed in the previous
paragraph. Fig. 10a clearly displays a decrease in thrust when the
blade is in the downward position, indicating that the presence of
the tower has a small but appreciable effect. This effect is most
marked at 5 m/s and decreases with speed to be completely masked
by uctuations at 25 m/s. Perceive that at 5 m/s CFD shows no
uctuations since the ow does not separate (see also Electronic
Annex I), while the experiments show signicant uctuations,
likely due to vibrations. The frequency spectrum at 5 m/s is
matched very well by CFD at low frequencies, showing a peak at
2.4 Hz. At higher wind speeds the lower frequencies are matched
reasonably but as expected higher frequencies are exhibiting
amplitudes much smaller than the experiments. Sezer-Uzol and
Long [15] computed the case with 15 m/s, showing uctuations of
the thrust coefcient of about 1.8%, compared with 9% of the NREL
experiments and 5% in this work.
Regarding performance of the two models, there are no major
differences between RANS and DES when wind velocity is small,
while small but clear differences can be seen at higher velocities on
time-averaged quantities. However, both models predict the same
trends and magnitudes for all blade sections, indicating that these
models are capable tools for wind turbine simulations, at least
within the conditions simulated herein. At higher wind speeds
(25 m/s), where separation is massive, DES appears to produce
better results, while the opposite occurs for lower wind speeds
(15 m/s). A more advanced DDES model [35] may be able to
improve on these results.
4.2. Variable pitch angle for constant wind speed
Computations with variable pitch follow one of the conditions of
the experimental sequence K. In this case the pitch angle is varied
from 15
to 40
in increments of 5
for wind speed of 15 m/s. The
resulting angles of attack at section r/R 0.30 vary from 48.2
to 1.6
and at r/R 0.95 vary from 34 to 16.3
. The experiments
were performed for a step up, then step down procedure, while the
computations were limited to the step up phase of the procedure.
In the experiments there was an 8 s hold between pitch steps, while
in CFD this hold was reduced to 4 s to save computational time.
Fig. 11 shows vortical structures represented by iso-surfaces of
Q 5, also shown in Electronic Annex IX in a blade-xed reference
system that provides greater insight of the ow eld as the blade
pitch angle is dynamically changed throughout the computation.
Fig. 12. Thrust for different pitch angles (wind velocity 15 m/s).
295
Fig. 13. Power to the shaft for different pitch angles (wind velocity 15 m/s).
At 15
of pitch the angle of attack is maximum and as expected the
separation is massive. The angle of attack is zero at the tip for
approximately 25
pitch, and at this condition there is a consequent
absence of tip vortices. At higher pitch angles the angle of attack
becomes negative and the turbine starts behaving as a fan. This can be
seen in Fig. 12, which shows the thrust force on the shaft as a function
of pitch angle. The thrust continuously decreases with increasing
pitch angle, and is negative for pitch angles bigger than 30
. Notice
the remarkably good predictions for all positive pitch angles, while
the difference with the experiments increases for very large angles of
attack (negative pitch angles), to reach 9% difference at 15
pitch.
The effect of pitch angle on power is shown in Fig. 13. This is an
important curve since it guides the design of the controller of
a pitch-controlled turbine. At wind speed of 15 m/s the maximum
power predicted by CFD is 19.1 kW and occurs at 15
pitch,
compared to the maximum experimental value of 18.9 kW at the
same pitch angle and a rated power of 19.8 kW. At this wind speed
the wind kinetic carries a potential of 160 kW, which implies
a maximum mechanical efciency of 12%. This value is particularly
low because under the cited conditions the turbine is in off-design
operation. At 15
pitch angle, the angle of attack is 21
at r/R 0.30
and 6.6
at r/R 0.95, while equal angles of attack would be ach-
ieved at wind speed of 7.5 m/s approximately. At high angles of
attack CFD tends to overpredict the power, while at higher blade
pitch angles (smaller and negative angles of attack) the trend
reverses and the power is underpredicted. Still the ability of the
code to predict this curve is remarkable.
The normal force coefcient Cn is shown in Fig. 14. The predictions
are in good agreement with experiments for a wide range of sectional
locations and pitch angles. Signicant underpredictions occur at high
Fig. 14. Radial normal force coefcients for different pitch angles (wind velocity 15 m/s).
296 Y. Li et al. / Renewable Energy 37 (2012) 285e298

Fig. 15. Pressure coefcients at different pitch angles and blade sections (wind speed 15 m/s, lines: CFD, symbols: experiments).
 Y. Li et al. / Renewable Energy 37 (2012) 285e298
Fig. 16. Dimensionless axial velocities at x/R 0.8, (wind speed 15 m/s).
angles of attack for r/R 0.47 and r/R 0.80, for moderate angles of
attack at r/R 0.63 and for low angle of attack at r/R 0.30. The
largest deviations are present at the innermost sections of the blade,
with excellent results for r/R 0.80 and r/R 0.95. Notice that the
normal force coefcient is, for the same blade pitch angle, mostly
larger in the inner sections of the blade, consistent with a larger angle
of attack in those sections. At r/R 0.95, where large ow separation
is observed up to about 10
of pitch (see Fig. 11), the normal force
coefcient remains at until the ow is attached, evidencing stall at
large angles of attack. The essentially at response with pitch angle
occurs 5w10
of pitch angle, and then the normal force decays faster
with decreased angle of attack, becoming negative for negative
angles of attack (pitch angles greater than 25
at r/R 0.95 and
greater than 38
at r/R 0.30).
Fig. 15 shows results of pressure distributions at three blade
sections for pitch angles from 15
to 40
. Predictions at r/R 0.95
are excellent in trend and magnitude and are very good at r/
R 0.63 with some magnitude issues discussed later. In contrast,
predictions at r/R 0.30 are excellent in trend and magnitude for
small angles of attack but fair for high angles of attack (pitch angles
from 15
to 10
). At section r/R 0.30 the experiments show
a much atter pressure coefcient on the suction side of the blade
than the CFD predictions, mostly for 10
of blade pitch angle where
the CFD and experiment discrepancies are largest. At 15
of pitch
angle and higher the experiments and CFD both show attached
ow and the agreement is excellent. The integral of the pressure on
the normal direction, leading to the normal force coefcient of
Fig. 14, is still excellent except for 10
of pitch angle, indicating that
the errors in pressure tend to cancel each other at other sections, as
can be seen in Fig. 14. At r/R 0.63 the predicted trends are in good
agreement with the data, showing stalled ow up to about 10
of
pitch angle, and attached ow thereafter. The magnitude of the
pressure on the suction side is a bit underpredicted for pitch angles
from 5
to 10
, resulting in underpredicted normal forces as seen
in Fig. 14. At section r/R 0.95 the ow is stalled with at pressure
distributions on the suction side up to 0
blade pitch, and then
remains mostly attached. The agreement at all angles of attack is
excellent. Notice that the suction and pressure sides are reversed
for negative angles of attack at all sections.
Cross-sections at x/R 0.8 showing instantaneous axial velocity
are depicted in Fig. 16, representing the near wakes at pitch
297
angles 15, 5, 5, 15, 25 and 40
. The axial velocity exhibits
decreases that are consistent with the level of power generated at
this wind speed, shown in Fig. 13. The vortical structures evident in
Fig. 11 at stall conditions (negative or small pitch angles) are
present in the form of large variations of axial velocity. For instance
at 15
pitch there are extensive areas with axial velocities higher
than the incoming wind speed, mainly near the tip of the blade, and
very low velocities for regions close to r/R 0.30, but again very
high wind speeds around the nacelle. On the other extreme, at 40

pitch the turbine is actually working as a fan and the axial velocities
are higher than the inlet velocity.
5. Conclusions
This study presents dynamic overset CFD simulations for the
NREL phase VI wind turbine. In particular, two sequences of the
experiment test cases are studied with complete turbine geometry,
including the NREL phase VI blades, and approximate geometries
for hub, nacelle and tower. RANS and DES models are used in the
simulations, and extensive comparisons with experimental data are
performed. The motion model coupled with the overset method-
ology allows for the presence of parent/children objects, enabling
the computation of variable blade pitch in a moving rotor with
respect to static tower, nacelle and ground. Results at constant pitch
and variable wind speed (experimental sequence S) or with
constant wind speed and variable pitch (experimental sequence K)
show that the CFD predictions match the experimental data
consistently well, including the general trends of power and thrust,
sectional normal force coefcients and pressure coefcients at
different sections along the blade. At very large angles of attack the
conditions are more demanding and the CFD results tend to slightly
overpredict the thrust and underpredict the power. Evaluation of
the transient pressure on the blades reveals that DES is able to
predict uctuations with similar frequencies to the experimental
measurements; however, at least at the level of grid resolution used
in this study, the amplitude is underpredicted, mainly at the
outermost sections.
Future work will focus on study of off-shore wind turbines in
oating structures in waves to take advantage of the air/water free
surface capabilities of CFDShip-Iowa. Fluid-structure interaction of
exible blades will also be studied.
298 Y. Li et al. / Renewable Energy 37 (2012) 285e298
Acknowledgments
Computations were performed at the National Center for
Computational Science in the Cray XT5 Jaguar, grant ARD005.
Experimental results were provided by Dr. Scott Schreck, from the
National Renewable Energy Laboratory at Boulder, CO. His help is
deeply appreciated.
References
[1] Archer CL. Evaluation of global wind power. Journal of Geophysical Research
2005;110:D12110.
[2] GWEC Global Wind Energy Council (GWEC). Global wind 2009 report; 2009.
[3] Glauert H. Airplane propellers. In: Durand WF, editor. Aerodynamic theory.
New York: Dover Publications; 1963.
[4] Shen WZ, Mikkelsen R, Srensen JN, Bak C. Tip loss corrections for wind
turbine computations. Wind Energy 2005;8:457e75.
[5] Leishman JG. A semi-empirical model for dynamic stall. Journal of the
American Helicopter Society 1989;34:3e17.
[6] Langtry RB, Gola J, Menter FR. Predicting 2D airfoil and 3D wind turbine rotor
performance using a transition model for general CFD codes. In: 44th AIAA
aerospace sciences meeting and exhibit, Reno, Nevada, 2006.
[7] Whale J, Fisichella CJ, Selig S. Correcting inow measurements from Hawts
using a lifting-surface code. In: Proceedings 1999 ASME wind energy
symposium, 1999. pp. 175e185.
[8] Hess JL. Review of integral-equation techniques for solving potential-ow
problems with emphasis on the surface-source method. Computer Methods
in Applied Mechanics and Engineering 1975;5:145e96.
[9] Landahl MT, Stark VJE. Numerical lifting-surface theory-problems and prog-
ress. AIAA Journal 1977;6:2049e60.
[10] Preuss RD, Morino L, Suciu EO. Unsteady potential aerodynamics of rotors
with applications to horizontal-axis windmills. AIAA Journal 1980;18:385e93.
[11] Johansen J, Srensen NN, Michelsen JA, Schreck S. Detached-eddy simulation
of ow around the NREL phase VI blade. Wind Energy 2002;5:185e97.
[12] Fingersh LJ, Sinuns D, Hand M, Jager D, Cotrell J, Robinson M, et al. Wind
tunnel testing of NRELs unsteady aerodynamics experiment. In: Proceedings
2001 ASME wind energy symposium, 2001. pp. 129e135.
[13] Hand MM, Simms DA, Fingersh LJ, Jager DW, Cotrell JR, Schreck S, et al.
Unsteady aerodynamics experiment phase VI: wind tunnel test congurations
and available data campaigns. National Renewable Energy Laboratory; 2001.
NREL/TP-500e29955.
[14] Schreck S. The NREL full-scale wind tunnel experiment. Introduction to the
special issue. Wind Energy 2002;5:77e84.
[15] Sezer-Uzol N, Long L., 3-D time-accurate CFD simulations of wind turbine
rotor ow elds. In: 44th AIAA aerospace sciences meeting and exhibit, Reno,
Nevada, 2006.
[16] Srensen NN, Michelsen JA, Schreck S. Navier-Stokes predictions of the NREL
phase VI rotor in the NASA Ames 80 ft  120 ft wind tunnel. Wind Energy
2002;5:151e69.
[17] Duque EPN, Burklund MD, Johnson W. Navier-Stokes and comprehensive
analysis performance predictions of the NREL phase VI experiment. Journal of
Solar Energy Engineering 2003;125:457e67.
[18] Buning PG, Parks SJ, Chan WM, Renze KJ. Application of the chimera over-
lapped grid scheme to simulation of space shuttle ascent ows. In: The 4th
international symposium on computational uid dynamics, Davis, California,
1991. pp. 132e137.
[19] Potsdam MA, Mavriplis DJ. Unstructured mesh CFD aerodynamic analysis of
the NREL phase VI rotor. In: 47th AIAA aerospace sciences meeting including
the New Horizons Forum and aerospace exposition, Orlando, Florida, 2009.
[20] Bazilevs Y, Hsu MC, Akkerman I, Wright S, Takizawa K, Henicke B, et al. 3D
simulation of wind turbine rotors at full scale. Part I: geometry modeling and
aerodynamics. International Journal for Numerical Methods in Fluids; 2010.
10.1002/d.2400.
[21] Jonkman J, Buttereld S, Musial W, Scott G. Denition of a 5-MW reference
wind turbine for offshore system development. National Renewable Energy
Laboratory; 2009.
[22] Huang J, Carrica PM, Stern F. Semi-coupled air/water immersed boundary
approach for curvilinear dynamic overset grids with application to ship
hydrodynamics. International Journal for Numerical Methods in Fluids 2008;
58:591e624.
[23] Carrica PM, Wilson RV, Stern F. An unsteady single-phase level set method for
viscous free surface ows. International Journal for Numerical Methods in
Fluids 2007;53:229e56.
[24] Carrica PM, Wilson RV, Noack RW, Stern F. Ship motions using single-phase
level set with dynamic overset grids. Computers & Fluids 2007;36:1415e33.
[25] Noack R. SUGGAR: a general capability for moving body overset grid
assembly. In: 17th AIAA computational uid dynamics conference, Toronto,
Ontario, Canada, 2005.
[26] Carrica PM, Huang J, Noack R, Kaushik D, Smith B, Stern F. Large-scale DES
computations of the forward speed diffraction and pitch and heave problems
for a surface combatant. Computers & Fluids 2010;39:1095e111.
[27] Paik K-J, Carrica PM, Lee D, Maki K. Strongly coupled uidestructure inter-
action method for structural loads on surface ships. Ocean Engineering 2009;
36:1346e57.
[28] Menter FR. Two-equation eddy-viscosity turbulence models for engineering
applications. AIAA Journal 1994;32:1598e605.
[29] Travin A, Shur M, Strelets M. Physical and numerical upgrades in the detached
eddy simulation of complex turbulent ows. In: Friedrich R, Rodi W, editors.
Advances in LES of complex ows; 2002. p. 239e54.
[30] Sainte-Rose B, Bertier N, Deck S, Dupoirieux F. A DES method applied to
a backward facing step reactive ow. Comptes Rendus Mcanique 2009;337:
340e51.
[31] Boger DA, Dreyer JJ. Prediction of hydrodynamic forces and moments for
underwater vehicles using overset grids. In: 44th AIAA aerospace sciences
meeting and exhibit, AIAA 2006-1148, Reno, Nevada, 2006.
[32] Hunt JCR, Wray AA, Moin P. Eddies, stream, and convergence zones in
turbulent ows. Center for Turbulence Research; 1988.
[33] Dousset V, Pothrat A. Formation mechanism of hairpin vortices in the wake
of a truncated square cylinder in a duct. Journal of Fluid Mechanics 2010;653:
519e36.
[34] Simms D, Schreck S, Hand M, Fingersh LJ. NREL unsteady aerodynamics
experiment in the NASA-Ames wind tunnel: a comparison of predictions to
measurements. National Renewable Energy Laboratory; 2001. NREL/TP-
500e29494.
[35] Shur M, Spalart P, Strelets M, Travin A. A hybrid RANS-LES approach with
delayed-DES and wall-modelled LES capabilities. International Journal of Heat
and Fluid Flow 2008;29:1638e49.